Unit 8: Introduction to Neural Networks. Perceptrons

Total Page:16

File Type:pdf, Size:1020Kb

Unit 8: Introduction to Neural Networks. Perceptrons Introduction Perceptrons Training with Delta rule Unit 8: Introduction to neural networks. Perceptrons D. Balbont´ınNoval F. J. Mart´ınMateos J. L. Ruiz Reina A. Riscos N´u~nez Departamento de Ciencias de la Computaci´one Inteligencia Artificial Universidad de Sevilla Inteligencia Artificial, 2012-13 Introduction Perceptrons Training with Delta rule Outline Introduction Perceptrons Training with Delta rule Introduction Perceptrons Training with Delta rule Artificial neurons: biological inspiration • Artificial neural networks are a computational model inspired on their biological counterpart • However, do not forget it is just a formal model: • Some features/properties of the biological systems are not captured in the computational model and viceversa • Our approach is to use them as a mathematical model that is the base of powerful automated learning algorithms • Structure: directed graph (nodes are artificial neurons) Introduction Perceptrons Training with Delta rule How a neural network works • Each node or unit (artificial neuron), is connected to other units via directed arcs (modeling the axon ! dendrites connexion) • Each arc j ! i propagates the output of unit j (denoted aj ) which in turn is one of the inputs for unit i. The inputs and outputs are numbers • Each arc j ! i has an associated numerical weight wji which determines the strength and the sign of the connexion (simulating a sinapsis) Introduction Perceptrons Training with Delta rule How a neural network works • Each unit calculates its output according to the received inputs • The output of each unit is, in turn, used as one of the inputs of other neurons • The calculation performed by each unit is very simple, as we will see later on • The network receives a series of external inputs (input units) and returns the output of some of its neurons, called output units Introduction Perceptrons Training with Delta rule Calculation at each unit Pn • The output of a unit is: ai = g( j=0 wji aj ) • Where: • g is the activation function Pn • The summatory j=0 wji aj (denoted by ini ) gathers all the outputs of units j connected with unit i • Except for j = 0, which is considered as a virtual input a0 = −1 and has a weight w0i called bias Introduction Perceptrons Training with Delta rule Bias and activation functions • Intuitively, the bias weight w0i of each unit is interpreted as the minimum amount that the sum of the received input signals has to reach in order to activate the unit • The role of the activation function g is to \normalize" the output (usually to 1) upon activation. Besides, this ingredient allows that the network offers something more than a simple lineal function • Frequently used activation functions: 1 si x > 0 • Bipolar: sgn(x) = −1 si x ≤ 0 1 si x > 0 • Threshold: threshold(x) = 0 si x ≤ 0 • 1 Sigmoid: σ(x) = 1+e−x • The sigmoid function is derivable and σ0(x) = σ(x)(1 − σ(x)) Introduction Perceptrons Training with Delta rule Feedforward neural networks • A feedforward neural network is an artificial neural network where connections between the units do not form a directed cycle (we focus on them in this unit) Introduction Perceptrons Training with Delta rule Feedforward neural networks • Units in a feedforward neural network are usually structured into layers, in such a way that each layer receives its inputs from units at the layer immediately before it • input layer, hidden layers and output layer • Multi-layer networks • Other architectures: recurrent networks, where the output units provide retroalimentation to the input units Introduction Perceptrons Training with Delta rule Neural networks as classifiers • A feedforward neural network with n units on the input layer and m units on the output layer computes a function from Rn into Rm • Thus, it can be used as a classifier for sets in Rn: • For Boolean classification, take m = 1 and: • If threshold or bipolar activation functions are used, then one of the outputs (e.g. 1) is interpreted as \YES" and the other one as \NO" • If sigmoid is used, then consider all output values over 0.5 as \YES" and any lower value as \NO" • In general, for classifiers with m possibles values, each output unit corresponds to a classification value; and then the unit having the higher output provides the classification Introduction Perceptrons Training with Delta rule Neural networks and Learning • Learning or training, in the framework of artificial neural networks, means searching adequate weights for the arcs, in order to get a desired behaviour (given by a training set) • More precisely, in feedforward neural networks, we usually follow the following supervised learning scheme • Given a training set n m D = f(x~d ; y~d ): x~d 2 R ; y~d 2 R ; d = 1;:::; kg • And given a structure of a network (number of layers and units per layer) n • Find a set of weight values wij such that the function from R into Rm represented by the network provides the best fit with the examples in the training set • We need a precise definition of \best fit” Introduction Perceptrons Training with Delta rule Practical applications of neural networks • For problems that can be expressed in terms of numbers (discrete or continuous) • Usually suitable for domains with a huge amount of data, possibly with noise: cameras, microphones, digitalized images, etc • We only care about the solution, not why it is so • Problems where we can afford a long training time for the network • And we want fast evaluation of new instances Introduction Perceptrons Training with Delta rule ALVINN: an example of an application • ANN trained to drive a car, at 70 Kms/h, according to the visual perception received as input from its sensors • Input of the network: The image of the road digitalized as an array of 30 × 32 pixels. That is, 960 input data • Output of the network: Indication about turning the wheel, encoded as a 30 component vector (from turn completely to the left, to keep straight, and then all the way to turn completely to the right) • Structure: feedforward network, input layer having 960 units, one hidden layer having 4 units and an output layer with 30 units Introduction Perceptrons Training with Delta rule ALVINN: an example of an application • Training: using a human driver, that drives the car again and again • The visual sensors record the image seen by the driver (sequences of 960 data each) • Other sensors record simultaneously the movements of the wheel • After properly encoding all the gathered info, we have a number of different pairs of the form (~x;~y), where ~x = (x1; x2;:::; x960) and ~y = (y1; y2;:::; y30), constitute examples of input/output for the network • Goal: find the best values for wji associated to arcs j ! i such that when the network receives ~x, its output matches the corresponding value ~y (or is the best possible approximation) Introduction Perceptrons Training with Delta rule Examples of practical applications • Classification • Pattern recognition • Optimization • Prediction: weather, audience, etc • Speech recognition • Artificial vision, image recognition • Constraint satisfaction • Control (robots, cars, etc) • Data compression • Diagnosis Introduction Perceptrons Training with Delta rule Outline Introduction Perceptrons Training with Delta rule Introduction Perceptrons Training with Delta rule Perceptrons • Let us first focus on the simplest case of neural network: feedforward, just one input layer and one output layer. • Since each of the output units is independent, without loss of generality we can consider just one unit in the output layer • This type of network is called perceptron • A perceptron using threshold activation function is able to represent the basic Boolean functions: Introduction Perceptrons Training with Delta rule Perceptrons: limitations • A perceptron with n input units, weights wi (i = 0;:::; n) and threshold (or bipolar) activation function, accepts those Pn (x1;:::; xn) such that i=0 wi xi > 0 (where x0 = −1) Pn n • The equation i=0 wi xi = 0 represents a hyperplane in R • That is, a Boolean function can only be represented by a threshold perceptron if there exists an hyperplane separating positive elements from negative elements (linearly separable) • Perceptrons with sigmoid activation function have similar expressive limitations Introduction Perceptrons Training with Delta rule Perceptrons: limitations • For example, functions AND and OR are linearly separable, while XOR is not: • Despite their expressive limitations, perceptrons have the advantage that there exists a simple training algorithm for perceptrons with threshold activation function • Able to find the suitable perceptron for any linearly separable training set Introduction Perceptrons Training with Delta rule Algorithm for (threshold) Perceptron training • Input: A training set D (with examples of the form (~x; y), with ~x 2 Rn and y 2 f0; 1g), and a learning factor η Algorithm 1) Consider randomly generated initial weights w~ (w0; w1;:::; wn) 2) Repeat until halting condition 1) For each (~x; y) in the training set do Pn 1) Calculate o = threshold( i=0 wi xi ) (with x0 = −1) 2) For each wi do: wi wi + η(y − o)xi 3) Return w~ Introduction Perceptrons Training with Delta rule Comments about the algorithm • η is a positive constant, usually very small (e.g. 0.1), called learning factor, that moderates the weights update process • At each iteration, if y = 1 and o = 0, then y − o = 1 > 0, and therefore wi corresponding to positive xi will increase (and those corresponding to negative xi will decrease). Thus, o (real output) will come closer
Recommended publications
  • CMPT 882 Machine Learning, 2004-1 Week 5 Lecture Notes Instructor
    CMPT 882 Machine Learning, 2004-1 Week 5 Lecture Notes Instructor: Dr. Oliver Schulte Scribe: Sarah Brown and Angie Zhang February 3rd and 5th Contents 1. Artificial Neural Network 1.1 Overview 1.2 Gradient Descent and the Delta Rule 1.2.1 Overview of Gradient Descent and the Delta Rule 1.2.2 Derivation of the Gradient Descent Rule 1.2.3 Stochastic Approximation to Gradient Descent 1.3 Multilayer Networks and the Backpropagation Algorithm 1.3.1 Multilayer Networks 1.3.2 The Backpropagation Algorithm 1.3.2.1 Adding Momentum 1.3.2.2 Learning in Arbitrary Acyclic Networks 1.3.2.3 Comments and examples on Backpropagation: 1.3.3 Derivation of the Backpropagation Rule 1.3.4 Efficiency 1.4 Convergence and Local Minima 1.5 Representational Power of Feedforward Networks 1.6 Hypothesis Space Search and Inductive Bias 1.7 Hidden Layer Representations 1.8 Generalization, Overfitting, and Stopping Criterion 1.9 Definitions 2 1. Artificial Neural Networks 1.1 Overview This section presumes familiarity with some basic concepts of Artificial Neural Network (ANN), since we have covered some of the materials on the previous lectures. This starts from Gradient Descent and the Delta Rule and continues to Backpropagation. In addition some of the benefits of the various techniques as well as the drawbacks are discussed. 1.2 Gradient Descent and the Delta Rule 1.2.1 Overview of Gradient Descent and the Delta Rule From the previous section, we know that the perceptron rule can find a successful weight vector when the training examples are linearly separable.
    [Show full text]
  • A Dynamical Study of the Generalised Delta Rule
    A Dynamical Study of the Generalised Delta Rule By Edward Butler, BA Thesis submitted to the University of Nottingham for the degree of Doctor of Philosophy, May, 2000 List of Contents INTRODUCTION ..................................................................................................................1 1.1 The Promise of the Neural Approach ...................................................................2 1.2 Early Linear Learning Algorithms ........................................................................3 THE GENERALISED DELTA RULE .................................................................................5 2.1 The Standard Generalised Delta Rule...................................................................7 2.2 Topologies ..........................................................................................................10 2.3 The Task Domain: Boolean Algebra ..................................................................14 2.4 Theoretical Problems..........................................................................................19 EMPIRICAL PROBLEMS..................................................................................................20 3.1 Selection of Training Regime .............................................................................20 3.2 Specification of Initial Conditions ......................................................................21 3.3 Selection of Learning Rate Parameter.................................................................23 3.4 The Problem
    [Show full text]
  • Artificial Neural Networks Supplement to 2001 Bioinformatics Lecture on Neural Nets
    Artificial Neural Networks Supplement to 2001 Bioinformatics Lecture on Neural Nets ಲং ྼႨ E-mail: [email protected] http://scai.snu.ac.kr./~btzhang/ Byoung-Tak Zhang School of Computer Science and Engineering SeoulNationalUniversity Outline 1. Basic Concepts of Neural Networks 2. Simple Perceptron and Delta Rule 3. Multilayer Perceptron and Backpropagation Learning 4. Applications of Neural Networks 5. Summary and Further Information 2 1. Basic Concepts of Neural Networks The Brain vs. Computer 1. 1011 neurons with 1. A single processor with 1014 synapses complex circuits 2. Speed: 10-3 sec 2. Speed: 10 –9 sec 3. Distributed processing 3. Central processing 4. Nonlinear processing 4. Arithmetic operation (linearity) 5. Parallel processing 5. Sequential processing 4 What Is a Neural Network? ! A new form of computing, inspired by biological (brain) models. ! A mathematical model composed of a large number of simple, highly interconnected processing elements. ! A computational model for studying learning and intelligence. 5 From Biological Neuron to Artificial Neuron 6 From Biology to Artificial Neural Networks (ANNs) 7 Properties of Artificial Neural Networks ! A network of artificial neurons ! Characteristics " Nonlinear I/O mapping " Adaptivity " Generalization ability " Fault-tolerance (graceful degradation) " Biological analogy <Multilayer Perceptron Network> 8 Synonyms for Neural Networks ! Neurocomputing ! Neuroinformatics (Neuroinformatik) ! Neural Information Processing Systems ! Connectionist Models ! Parallel Distributed Processing (PDP) Models ! Self-organizing Systems ! Neuromorphic Systems 9 Brief History ! William James (1890): Describes (in words and figures) simple distributed networks and Hebbian Learning ! McCulloch & Pitts (1943): Binary threshold units that perform logical operations (they prove universal computations!) ! Hebb (1949): Formulation of a physiological (local) learning rule ! Rosenblatt (1958): The Perceptron - a first real learning machine ! Widrow & Hoff (1960): ADALINE and the Windrow-Hoff supervised learning rule.
    [Show full text]
  • Artificial Neural Networks 2Nd February 2017, Aravindh Mahendran, Student D.Phil in Engineering Science, University of Oxford
    Artificial Neural Networks 2nd February 2017, Aravindh Mahendran, Student D.Phil in Engineering Science, University of Oxford Contents Introduction ............................................................................................................................................ 1 Perceptron .............................................................................................................................................. 2 LMS / Delta Rule for Learning a Perceptron model ............................................................................ 4 Demo for Simple Synthetic Dataset .................................................................................................... 4 Problems ............................................................................................................................................. 6 Stochastic Approximation of the Gradient ......................................................................................... 6 Sigmoid Unit ............................................................................................................................................ 6 Backpropagation algorithm ................................................................................................................ 7 Backpropagation Algorithm for Vector Input and Vector Output ...................................................... 9 Problems ............................................................................................................................................
    [Show full text]
  • Artificial Neural Networks
    Artificial Neural Networks 鮑興國 Ph.D. National Taiwan University of Science and Technology Outline Perceptrons Gradient descent Multi-layer networks Backpropagation Hidden layer representations Examples Advanced topics What is an Artificial Neural Network? It is a formalism for representing functions inspired from biological learning systems The network is composed of parallel computing units which each computes a simple function Some useful computations taking place in Feedforward Multilayer Neural Networks are Summation Multiplication Threshold (e.g., 1/(1 + e-x), the sigmoidal threshold function). Other functions are also possible Biological Motivation dendrites cell axon synapse dendrites Biological Learning Systems are built of very complex webs of interconnected neurons Information-Processing abilities of biological neural systems must follow from highly parallel processes operating on representations that are distributed over many neurons ANNs attempt to capture this mode of computation Biological Neural Systems Neuron switching time : > 10-3 secs Computer takes 10-10 secs Number of neurons in the human brain: ~1011 Connections (synapses) per neuron: ~104-105 Face recognition : ~0.1 secs 100 inference steps? Brain must be parallel! High degree of parallel computation Distributed representations Properties of Artificial Neural Nets (ANNs) Many simple neuron-like threshold switching units Many weighted interconnections among units Highly parallel, distributed processing Learning by tuning the connection weights ANNs are motivated by biological neural systems; but not as complex as biological systems For instance, individual units in ANN output a single constant value instead of a complex time series of spikes A Brief History of Neural Networks (Pomerleau) 1943: McCulloch and Pitts proposed a model of a neuron → Perceptron (Mitchell, section 4.4) 1960s: Widrow and Hoff explored Perceptron networks (which they called “Adelines”) and the delta rule.
    [Show full text]
  • Chapter 6 Convolutional Neural Network
    Neural Network 生醫光電所 吳育德 Phil Kim, MATLAB Deep Learning With Machine Learning, Neural Networks and Artificial Intelligence, 2017, chapter 2. Neural Network • The models of Machine Learning can be implemented in various forms. The neural network is one of them. x1, x2, and x3 are the input signals. w1, w2, and w3 are the weights for the corresponding signals. b is the bias. A node that receives three inputs. • The circle and arrow of the figure denote the node and signal flow, respectively. • The information of the neural net is stored in the form of weights and bias. The equation of the weighted sum can be written with matrices as: (Equation 2.1) Where w and x are defined as: Finally, the node enters the weighted sum into the activation function and yields its output: Layers of Neural Network • The group of square nodes in figure is called the input layer. They do not calculate the weighted sum and activation function. • The group of the rightmost nodes is called the output layer. The output from these nodes becomes the final result of the neural network. • The layers in between the input and output layers are called hidden layers. Example A neural network with a single hidden layer. The activation function of each node is a linear function. The first node of the hidden layer calculates the output as: In a similar manner, the second node of the hidden layer calculates the output as: Matrix equation Supervised Learning of a Neural Network 1. Initialize the weights with adequate values. 2. Take the “input” from the training data { input, correct output }, and enter it into the neural network.
    [Show full text]
  • Backpropagation Generalized Delta Rule for the Selective Attention Sigma–If Artificial Neural Network
    Int. J. Appl. Math. Comput. Sci., 2012, Vol. 22, No. 2, 449–459 DOI: 10.2478/v10006-012-0034-5 BACKPROPAGATION GENERALIZED DELTA RULE FOR THE SELECTIVE ATTENTION SIGMA–IF ARTIFICIAL NEURAL NETWORK MACIEJ HUK Institute of Informatics Wrocław University of Technology, Wyb. Wyspia´nskiego 27, 50-370 Wrocław, Poland e-mail: [email protected] In this paper the Sigma-if artificial neural network model is considered, which is a generalization of an MLP network with sigmoidal neurons. It was found to be a potentially universal tool for automatic creation of distributed classification and selective attention systems. To overcome the high nonlinearity of the aggregation function of Sigma-if neurons, the training process of the Sigma-if network combines an error backpropagation algorithm with the self-consistency paradigm widely used in physics. But for the same reason, the classical backpropagation delta rule for the MLP network cannot be used. The general equation for the backpropagation generalized delta rule for the Sigma-if neural network is derived and a selection of experimental results that confirm its usefulness are presented. Keywords: artificial neural networks, selective attention, self consistency, error backpropagation, delta rule. 1. Introduction ing artificial neural networks. Unfortunately, networks that use higher-order neuron models, such as Sigma-Pi In nature, selective attention is a mechanism which pro- (Feldman and Ballard, 1982; Rumelhart et al., 1986; Mel, vides living organisms with the possibility to sift incom- 1990; Olshausen et al., 1993), Power Unit (Durbin and ing data to extract information which is most important Rumelhart, 1990) or Clusteron (Mel, 1992), realize only at a given moment and which should be processed in a very limited set of attentional mechanisms (Neville and detail (Broadbent, 1982; Treisman, 1960).
    [Show full text]